graphics driver architectural design flawed?

[_______0]

I see a problem with the way graphics drivers improvements are added to linux. For instance new features and kewl stuff are added to not release kernel yet and older kernel get excluded.

Then in order to use some new graphics capabilities there's no option but to either wait until the WHOLE kernel is stabilized or to use unstable sources. Seems asstarded.

In m$$ one can use latest drivers with same install, no need to change m$$$ versions.

Why isn't graphics drivers version agnostic, a module that works accross ALL 3.X.X versions?? Seems idiotic to attach an opengl feature implementation just for a newer kernel.

Thanks.

[Ant P.]

This problem is easy to fix, just use `modprobe timemachine` to load hardware drivers from the future, or revert to XFree86 4.4.0 to experience the well-thought-out userspace driver architecture you're proposing.

[Hu]

It is well accepted that if you want new features, you must use a new kernel. Large features typically are not backported, though users are welcome to attempt it. Your claim that Windows permits the use of newer drivers with an existing install is misinformed. Microsoft routinely changes the driver model, so graphics drivers designed for Windows 8 will not work with Windows XP. Yet according to your argument, this should be supported, so that users can stay on a stable Windows XP system instead of being subjected to Microsoft's latest experiment, but still enjoy the latest hardware support.

[John R. Graham]

I think the other point is those older stable kernels probably wouldn't be as stable if all those changes were backported to them.

- John
_________________
I can confirm that I have received between 0 and 499 National Security Letters.

[_______0]

adding a single feature such as:

GLX_MESA_query_renderer

(from phoronix news) would make the entire graphics stack so immensely unstable? As far as I understand these opengl features are completely independent of lowlevel kernel changes.

OpenGL features trickle down at a very low pace into open source drivers (wut, amd no video accel yet?), such as it makes waiting for each an excruciating agony.

I would love something like this:

new extension -> new version gfx kernel module ebuild -> emerge -> rmm && modprobe -> done.

Can't be that hard and THAT desestabilizing.

[Hu]

[Gusar]

AMD video accel has absolutely nothing to do with OpenGL.

As for the general issue you have, there's compat-drivers.